搭建卷积

x=torch.randn(1, 1, 30, 30)

weight=torch.Tensor(

             3, 1, *(3,3))

b=torch.Tensor(

             1)

print(F.conv2d(x,weight,stride=1,padding=0).shape)

class CondConv2D(nn.Module):

    def __init__(self, in_channels, out_channels, kernel_size, stride=1,

                 padding=0, dilation=1):

        super(CondConv2D,self).__init__()

        self.weight=Parameter(torch.Tensor(

              out_channels, in_channels, *(3,3))

                              )    

    def forward(self, inputs):

        return F.conv2d(inputs,self.weight,stride=1,padding=0)